What is the Value of Censored Data? An Exact Analysis for the Data-driven Newsvendor

Information Systems and Operations Management

Speaker: Omar Mouchtaki (NYU Stern School of Business)

Room Bernard Ramanantsoa

Abstract: 

We study the newsvendor problem with historical sales data, which reveal demand only up to available inventory. Departing from concentration-based arguments, we characterize the exact worst-case performance of standard data-driven policies for any sample size and censoring level. Our central result reduces the infinite-dimensional problem of evaluating a policy’s worst-case performance over demand distributions to a finite-dimensional optimization problem. This yields the first finite-sample guarantee for the standard Kaplan–Meier inventory policy and precisely quantifies the value of censored information. We further show how this reduction can guide the design of new policies under censoring.

Bio: 

Omar Mouchtaki is an Assistant Professor at NYU Stern School of Business. His research examines how to make operational decisions with limited, heterogeneous or incomplete data. He develops methods that connect statistical learning with optimization, with applications to inventory management, pricing and auctions. His recent works also study how artificial intelligence can support human decision-making.